Park, So Jin Yoon, Hyo In Lee, HyeIn Kim, Min-Chae Yang, Jung, Dae-Hyun Ahn, Ju Yeon Park, Soo Hyun 2024-07-26T06:30:23Z 2024-07-26T06:30:23Z 2024-07-26 2024-06 1738-1266 https://pubs.kist.re.kr/handle/201004/150308 Purpose Processed herbal medicines in East Asia are often misused or mixed because distinguishing their different forms is difficult using the naked eye. However, an artificial intelligence model can address this problem by classifying herbal medicines with similar shapes. Methods This study proposes using hyperspectral images to classify herbal medicines from five plant species belonging to the Apiaceae family: Angelica Gigas, Angelica acutiloba, Saposhnikovia divaricata, Peucedanum japonicum, and Glehnia littoralis. To classify the hyperspectral images, we evaluated the performance using six machine learning-based models―SVM, KNN, DT, RF, LR, and LightGBM; preprocessed the original data with a Savitzky-Golay filter, first derivative, and second derivative; and evaluated processing changes in accuracy. Results The logistic regression model with the second derivative showed the best performance, with an accuracy of 0.99. The classification prediction error was due to the very high similarity between the second derivative data of the Peucedani Japonici Radix and Glehnia littoralis. The above results confirmed that the second derivative during preprocessing and logistic regression model performed best when analyzing the hyperspectral images. Conclusion Hyperspectral data is proposed for the hyperspectral image analysis of herbal medicines that use the root zone as a medicinal site, owing to its excellent performance. English 한국농업기계학회 Evaluating the Accuracy of Machine Learning Classification Models for Similar Herbal Medicine Using Hyperspectral Imaging Article 10.1007/s42853-024-00224-1 1 Journal of Biosystems Engineering, v.49, no.2, pp.156 - 166 Journal of Biosystems Engineering 49 2 156 166 N scopus kci ART003099676 2-s2.0-85193812842 Root medicine · Phenotyping · Apiaceae · Data preprocessing · Logistic regression